Evaluating the Classification Potential of Eye-Tracking Measures Based on Perception of Social and Physical Contingencies in Toddlers with ASD
Objectives: The goal of this study is to understand how physical and social contingencies bias attention in ASD and TD toddlers, and to evaluate the potential of eye-tracking measures of behavioral responses to social and physical contingencies as biomarkers for discriminating between toddlers with ASD and TD peers.
Methods: TD (N=22) and ASD (N=30) toddlers (Mean ± S.D: 23.3 ± 7.3 months) participated in a preferential looking paradigm, presenting audiovisual stimuli that varied in social context – circles, mouth-like ellipses and faces paired with tones or speech – to calibrate the biasing effect of social context and audiovisual synchrony on visual attention. To investigate generalization to naturalistic settings, a second cohort of TD (N=23) and ASD (N=44) toddlers (23.5 ± 6.1 months) were presented with videos showing a caregiver and a moving toy synchronized with the caregiver’s speech against a playroom background. Eye-tracking measures of relative visual fixation on face, object, and background were collected and used to derive optimal classifiers for discriminating ASD from TD participants. Performance was evaluated by constructing Receiver Operating Characteristics (ROC) and calculating the sensitivity, specificity, and area under the curve (AUC) with cross-validation.
Results: When infants were presented with split-screen audiovisual stimuli consisting of caregiver faces and circles synchronized to the caregiver’s speech, percentage of total fixation time spent on faces was significantly higher (P<.001) for TD controls (79.8 ± 21.2%) relative to ASD peers (56.2 ± 30.3%). Using this measure to classify TD and ASD toddlers yielded an ROC with sensitivity 71%, specificity 72%, and AUC = 0.78. When presented with naturalistic stimuli, percentage of total fixation time spent on objects was significantly higher (P<.001) for ASD toddlers (34.1 ± 23.9%) compared to TD peers (17.9 ± 19.3%). Using this measure yielded an ROC with sensitivity 79%, specificity 79%, and AUC = 0.86.
Conclusions: Visual scanning patterns of toddlers with ASD are biased by the presence of social and physical contingencies in ways that differentiate them from their TD peers, to the extent that eye-tracking measures of visual fixation can successfully discriminate between diagnostic groups. In naturalistic settings, the presence of physical contingencies draws ASD toddlers’ attention away from relevant social context, suggesting a mechanism of derailment of normative developmental processes of learning.